refactor(ops): migrate rearrange to canonical InfiniOps API

[voltjia]

Summary

• migrate the InfiniCore rearrange adapter from deprecated RearrangeInfinilm to canonical Copy
• pass non_blocking=false, matching the legacy synchronous same-device behavior
• keep the InfiniCore public API, graph planning, and active dispatcher selection unchanged

API alignment

InfiniCore adapter	Previous InfiniOps API	Canonical InfiniOps API	Alignment target and evidence
Rearrange	RearrangeInfinilm::Call(handle, config, src, out)	Copy::Call(handle, config, src, false, out)	PyTorch Tensor.copy_(src, non_blocking=False). InfiniOps preserves src, non_blocking and represents the mutable receiver as the trailing explicit out tensor per its input-attribute-output convention.

Dependency

This PR is based on refactor/migrate-paged-caching-infiniops / #1467 because that branch advances the InfiniOps gitlink to e733e325b4af776261419e4dc7e1f9386c05fa71, which includes the canonical Copy operator from InfiniOps #872.

Against #1467, this PR changes one file with three insertions and three deletions. After #1467 merges, this PR can be retargeted to main without carrying an additional submodule update.

Validation

Validated commit e89ff869 remotely on ssh nvidia with accelerator-dev/nvidia:latest, using the exact InfiniOps gitlink e733e325:

• built and installed _infinicore with the canonical copy provider in the wrapper allowlist
• compiled and ran a temporary C++ smoke test that directly invoked rearrange_impl::infiniops::plan, run, and cleanup on an NVIDIA tensor; copied values and plan cleanup were verified
• clang-format 16.0.6 --dry-run --Werror passed for the changed file
• git diff --check passed

[voltjia]

Superseded by #1480, which consolidates the canonical InfiniOps adapter migrations while preserving each logical change as a separate commit.